2 * linux/net/sunrpc/xprt.c
4 * This is a generic RPC call interface supporting congestion avoidance,
5 * and asynchronous calls.
7 * The interface works like this:
9 * - When a process places a call, it allocates a request slot if
10 * one is available. Otherwise, it sleeps on the backlog queue
12 * - Next, the caller puts together the RPC message, stuffs it into
13 * the request struct, and calls xprt_transmit().
14 * - xprt_transmit sends the message and installs the caller on the
15 * transport's wait list. At the same time, it installs a timer that
16 * is run after the packet's timeout has expired.
17 * - When a packet arrives, the data_ready handler walks the list of
18 * pending requests for that transport. If a matching XID is found, the
19 * caller is woken up, and the timer removed.
20 * - When no reply arrives within the timeout interval, the timer is
21 * fired by the kernel and runs xprt_timer(). It either adjusts the
22 * timeout values (minor timeout) or wakes up the caller with a status
24 * - When the caller receives a notification from RPC that a reply arrived,
25 * it should release the RPC slot, and process the reply.
26 * If the call timed out, it may choose to retry the operation by
27 * adjusting the initial timeout value, and simply calling rpc_call
30 * Support for async RPC is done through a set of RPC-specific scheduling
31 * primitives that `transparently' work for processes as well as async
32 * tasks that rely on callbacks.
34 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
36 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
39 #include <linux/module.h>
41 #include <linux/types.h>
42 #include <linux/interrupt.h>
43 #include <linux/workqueue.h>
44 #include <linux/net.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/metrics.h>
54 # define RPCDBG_FACILITY RPCDBG_XPRT
60 static void xprt_request_init(struct rpc_task
*, struct rpc_xprt
*);
61 static inline void do_xprt_reserve(struct rpc_task
*);
62 static void xprt_connect_status(struct rpc_task
*task
);
63 static int __xprt_get_cong(struct rpc_xprt
*, struct rpc_task
*);
66 * The transport code maintains an estimate on the maximum number of out-
67 * standing RPC requests, using a smoothed version of the congestion
68 * avoidance implemented in 44BSD. This is basically the Van Jacobson
69 * congestion algorithm: If a retransmit occurs, the congestion window is
70 * halved; otherwise, it is incremented by 1/cwnd when
72 * - a reply is received and
73 * - a full number of requests are outstanding and
74 * - the congestion window hasn't been updated recently.
76 #define RPC_CWNDSHIFT (8U)
77 #define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
78 #define RPC_INITCWND RPC_CWNDSCALE
79 #define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
81 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
84 * xprt_reserve_xprt - serialize write access to transports
85 * @task: task that is requesting access to the transport
87 * This prevents mixing the payload of separate requests, and prevents
88 * transport connects from colliding with writes. No congestion control
91 int xprt_reserve_xprt(struct rpc_task
*task
)
93 struct rpc_xprt
*xprt
= task
->tk_xprt
;
94 struct rpc_rqst
*req
= task
->tk_rqstp
;
96 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
97 if (task
== xprt
->snd_task
)
103 xprt
->snd_task
= task
;
105 req
->rq_bytes_sent
= 0;
111 dprintk("RPC: %5u failed to lock transport %p\n",
113 task
->tk_timeout
= 0;
114 task
->tk_status
= -EAGAIN
;
115 if (req
&& req
->rq_ntrans
)
116 rpc_sleep_on(&xprt
->resend
, task
, NULL
, NULL
);
118 rpc_sleep_on(&xprt
->sending
, task
, NULL
, NULL
);
121 EXPORT_SYMBOL_GPL(xprt_reserve_xprt
);
123 static void xprt_clear_locked(struct rpc_xprt
*xprt
)
125 xprt
->snd_task
= NULL
;
126 if (!test_bit(XPRT_CLOSE_WAIT
, &xprt
->state
) || xprt
->shutdown
) {
127 smp_mb__before_clear_bit();
128 clear_bit(XPRT_LOCKED
, &xprt
->state
);
129 smp_mb__after_clear_bit();
131 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
135 * xprt_reserve_xprt_cong - serialize write access to transports
136 * @task: task that is requesting access to the transport
138 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
139 * integrated into the decision of whether a request is allowed to be
140 * woken up and given access to the transport.
142 int xprt_reserve_xprt_cong(struct rpc_task
*task
)
144 struct rpc_xprt
*xprt
= task
->tk_xprt
;
145 struct rpc_rqst
*req
= task
->tk_rqstp
;
147 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
148 if (task
== xprt
->snd_task
)
152 if (__xprt_get_cong(xprt
, task
)) {
153 xprt
->snd_task
= task
;
155 req
->rq_bytes_sent
= 0;
160 xprt_clear_locked(xprt
);
162 dprintk("RPC: %5u failed to lock transport %p\n", task
->tk_pid
, xprt
);
163 task
->tk_timeout
= 0;
164 task
->tk_status
= -EAGAIN
;
165 if (req
&& req
->rq_ntrans
)
166 rpc_sleep_on(&xprt
->resend
, task
, NULL
, NULL
);
168 rpc_sleep_on(&xprt
->sending
, task
, NULL
, NULL
);
171 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong
);
173 static inline int xprt_lock_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
177 spin_lock_bh(&xprt
->transport_lock
);
178 retval
= xprt
->ops
->reserve_xprt(task
);
179 spin_unlock_bh(&xprt
->transport_lock
);
183 static void __xprt_lock_write_next(struct rpc_xprt
*xprt
)
185 struct rpc_task
*task
;
186 struct rpc_rqst
*req
;
188 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
191 task
= rpc_wake_up_next(&xprt
->resend
);
193 task
= rpc_wake_up_next(&xprt
->sending
);
198 req
= task
->tk_rqstp
;
199 xprt
->snd_task
= task
;
201 req
->rq_bytes_sent
= 0;
207 xprt_clear_locked(xprt
);
210 static void __xprt_lock_write_next_cong(struct rpc_xprt
*xprt
)
212 struct rpc_task
*task
;
214 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
216 if (RPCXPRT_CONGESTED(xprt
))
218 task
= rpc_wake_up_next(&xprt
->resend
);
220 task
= rpc_wake_up_next(&xprt
->sending
);
224 if (__xprt_get_cong(xprt
, task
)) {
225 struct rpc_rqst
*req
= task
->tk_rqstp
;
226 xprt
->snd_task
= task
;
228 req
->rq_bytes_sent
= 0;
234 xprt_clear_locked(xprt
);
238 * xprt_release_xprt - allow other requests to use a transport
239 * @xprt: transport with other tasks potentially waiting
240 * @task: task that is releasing access to the transport
242 * Note that "task" can be NULL. No congestion control is provided.
244 void xprt_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
246 if (xprt
->snd_task
== task
) {
247 xprt_clear_locked(xprt
);
248 __xprt_lock_write_next(xprt
);
251 EXPORT_SYMBOL_GPL(xprt_release_xprt
);
254 * xprt_release_xprt_cong - allow other requests to use a transport
255 * @xprt: transport with other tasks potentially waiting
256 * @task: task that is releasing access to the transport
258 * Note that "task" can be NULL. Another task is awoken to use the
259 * transport if the transport's congestion window allows it.
261 void xprt_release_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
263 if (xprt
->snd_task
== task
) {
264 xprt_clear_locked(xprt
);
265 __xprt_lock_write_next_cong(xprt
);
268 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong
);
270 static inline void xprt_release_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
272 spin_lock_bh(&xprt
->transport_lock
);
273 xprt
->ops
->release_xprt(xprt
, task
);
274 spin_unlock_bh(&xprt
->transport_lock
);
278 * Van Jacobson congestion avoidance. Check if the congestion window
279 * overflowed. Put the task to sleep if this is the case.
282 __xprt_get_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
284 struct rpc_rqst
*req
= task
->tk_rqstp
;
288 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
289 task
->tk_pid
, xprt
->cong
, xprt
->cwnd
);
290 if (RPCXPRT_CONGESTED(xprt
))
293 xprt
->cong
+= RPC_CWNDSCALE
;
298 * Adjust the congestion window, and wake up the next task
299 * that has been sleeping due to congestion
302 __xprt_put_cong(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
307 xprt
->cong
-= RPC_CWNDSCALE
;
308 __xprt_lock_write_next_cong(xprt
);
312 * xprt_release_rqst_cong - housekeeping when request is complete
313 * @task: RPC request that recently completed
315 * Useful for transports that require congestion control.
317 void xprt_release_rqst_cong(struct rpc_task
*task
)
319 __xprt_put_cong(task
->tk_xprt
, task
->tk_rqstp
);
321 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong
);
324 * xprt_adjust_cwnd - adjust transport congestion window
325 * @task: recently completed RPC request used to adjust window
326 * @result: result code of completed RPC request
328 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
330 void xprt_adjust_cwnd(struct rpc_task
*task
, int result
)
332 struct rpc_rqst
*req
= task
->tk_rqstp
;
333 struct rpc_xprt
*xprt
= task
->tk_xprt
;
334 unsigned long cwnd
= xprt
->cwnd
;
336 if (result
>= 0 && cwnd
<= xprt
->cong
) {
337 /* The (cwnd >> 1) term makes sure
338 * the result gets rounded properly. */
339 cwnd
+= (RPC_CWNDSCALE
* RPC_CWNDSCALE
+ (cwnd
>> 1)) / cwnd
;
340 if (cwnd
> RPC_MAXCWND(xprt
))
341 cwnd
= RPC_MAXCWND(xprt
);
342 __xprt_lock_write_next_cong(xprt
);
343 } else if (result
== -ETIMEDOUT
) {
345 if (cwnd
< RPC_CWNDSCALE
)
346 cwnd
= RPC_CWNDSCALE
;
348 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
349 xprt
->cong
, xprt
->cwnd
, cwnd
);
351 __xprt_put_cong(xprt
, req
);
353 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd
);
356 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
357 * @xprt: transport with waiting tasks
358 * @status: result code to plant in each task before waking it
361 void xprt_wake_pending_tasks(struct rpc_xprt
*xprt
, int status
)
364 rpc_wake_up_status(&xprt
->pending
, status
);
366 rpc_wake_up(&xprt
->pending
);
368 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks
);
371 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
372 * @task: task to be put to sleep
375 void xprt_wait_for_buffer_space(struct rpc_task
*task
)
377 struct rpc_rqst
*req
= task
->tk_rqstp
;
378 struct rpc_xprt
*xprt
= req
->rq_xprt
;
380 task
->tk_timeout
= req
->rq_timeout
;
381 rpc_sleep_on(&xprt
->pending
, task
, NULL
, NULL
);
383 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space
);
386 * xprt_write_space - wake the task waiting for transport output buffer space
387 * @xprt: transport with waiting tasks
389 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
391 void xprt_write_space(struct rpc_xprt
*xprt
)
393 if (unlikely(xprt
->shutdown
))
396 spin_lock_bh(&xprt
->transport_lock
);
397 if (xprt
->snd_task
) {
398 dprintk("RPC: write space: waking waiting task on "
400 rpc_wake_up_task(xprt
->snd_task
);
402 spin_unlock_bh(&xprt
->transport_lock
);
404 EXPORT_SYMBOL_GPL(xprt_write_space
);
407 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
408 * @task: task whose timeout is to be set
410 * Set a request's retransmit timeout based on the transport's
411 * default timeout parameters. Used by transports that don't adjust
412 * the retransmit timeout based on round-trip time estimation.
414 void xprt_set_retrans_timeout_def(struct rpc_task
*task
)
416 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
418 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def
);
421 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
422 * @task: task whose timeout is to be set
424 * Set a request's retransmit timeout using the RTT estimator.
426 void xprt_set_retrans_timeout_rtt(struct rpc_task
*task
)
428 int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
429 struct rpc_rtt
*rtt
= task
->tk_client
->cl_rtt
;
430 struct rpc_rqst
*req
= task
->tk_rqstp
;
431 unsigned long max_timeout
= req
->rq_xprt
->timeout
.to_maxval
;
433 task
->tk_timeout
= rpc_calc_rto(rtt
, timer
);
434 task
->tk_timeout
<<= rpc_ntimeo(rtt
, timer
) + req
->rq_retries
;
435 if (task
->tk_timeout
> max_timeout
|| task
->tk_timeout
== 0)
436 task
->tk_timeout
= max_timeout
;
438 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt
);
440 static void xprt_reset_majortimeo(struct rpc_rqst
*req
)
442 struct rpc_timeout
*to
= &req
->rq_xprt
->timeout
;
444 req
->rq_majortimeo
= req
->rq_timeout
;
445 if (to
->to_exponential
)
446 req
->rq_majortimeo
<<= to
->to_retries
;
448 req
->rq_majortimeo
+= to
->to_increment
* to
->to_retries
;
449 if (req
->rq_majortimeo
> to
->to_maxval
|| req
->rq_majortimeo
== 0)
450 req
->rq_majortimeo
= to
->to_maxval
;
451 req
->rq_majortimeo
+= jiffies
;
455 * xprt_adjust_timeout - adjust timeout values for next retransmit
456 * @req: RPC request containing parameters to use for the adjustment
459 int xprt_adjust_timeout(struct rpc_rqst
*req
)
461 struct rpc_xprt
*xprt
= req
->rq_xprt
;
462 struct rpc_timeout
*to
= &xprt
->timeout
;
465 if (time_before(jiffies
, req
->rq_majortimeo
)) {
466 if (to
->to_exponential
)
467 req
->rq_timeout
<<= 1;
469 req
->rq_timeout
+= to
->to_increment
;
470 if (to
->to_maxval
&& req
->rq_timeout
>= to
->to_maxval
)
471 req
->rq_timeout
= to
->to_maxval
;
474 req
->rq_timeout
= to
->to_initval
;
476 xprt_reset_majortimeo(req
);
477 /* Reset the RTT counters == "slow start" */
478 spin_lock_bh(&xprt
->transport_lock
);
479 rpc_init_rtt(req
->rq_task
->tk_client
->cl_rtt
, to
->to_initval
);
480 spin_unlock_bh(&xprt
->transport_lock
);
484 if (req
->rq_timeout
== 0) {
485 printk(KERN_WARNING
"xprt_adjust_timeout: rq_timeout = 0!\n");
486 req
->rq_timeout
= 5 * HZ
;
491 static void xprt_autoclose(struct work_struct
*work
)
493 struct rpc_xprt
*xprt
=
494 container_of(work
, struct rpc_xprt
, task_cleanup
);
496 xprt_disconnect(xprt
);
497 xprt
->ops
->close(xprt
);
498 xprt_release_write(xprt
, NULL
);
502 * xprt_disconnect - mark a transport as disconnected
503 * @xprt: transport to flag for disconnect
506 void xprt_disconnect(struct rpc_xprt
*xprt
)
508 dprintk("RPC: disconnected transport %p\n", xprt
);
509 spin_lock_bh(&xprt
->transport_lock
);
510 xprt_clear_connected(xprt
);
511 xprt_wake_pending_tasks(xprt
, -ENOTCONN
);
512 spin_unlock_bh(&xprt
->transport_lock
);
514 EXPORT_SYMBOL_GPL(xprt_disconnect
);
517 xprt_init_autodisconnect(unsigned long data
)
519 struct rpc_xprt
*xprt
= (struct rpc_xprt
*)data
;
521 spin_lock(&xprt
->transport_lock
);
522 if (!list_empty(&xprt
->recv
) || xprt
->shutdown
)
524 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
526 spin_unlock(&xprt
->transport_lock
);
527 if (xprt_connecting(xprt
))
528 xprt_release_write(xprt
, NULL
);
530 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
533 spin_unlock(&xprt
->transport_lock
);
537 * xprt_connect - schedule a transport connect operation
538 * @task: RPC task that is requesting the connect
541 void xprt_connect(struct rpc_task
*task
)
543 struct rpc_xprt
*xprt
= task
->tk_xprt
;
545 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task
->tk_pid
,
546 xprt
, (xprt_connected(xprt
) ? "is" : "is not"));
548 if (!xprt_bound(xprt
)) {
549 task
->tk_status
= -EIO
;
552 if (!xprt_lock_write(xprt
, task
))
554 if (xprt_connected(xprt
))
555 xprt_release_write(xprt
, task
);
558 task
->tk_rqstp
->rq_bytes_sent
= 0;
560 task
->tk_timeout
= xprt
->connect_timeout
;
561 rpc_sleep_on(&xprt
->pending
, task
, xprt_connect_status
, NULL
);
562 xprt
->stat
.connect_start
= jiffies
;
563 xprt
->ops
->connect(task
);
568 static void xprt_connect_status(struct rpc_task
*task
)
570 struct rpc_xprt
*xprt
= task
->tk_xprt
;
572 if (task
->tk_status
>= 0) {
573 xprt
->stat
.connect_count
++;
574 xprt
->stat
.connect_time
+= (long)jiffies
- xprt
->stat
.connect_start
;
575 dprintk("RPC: %5u xprt_connect_status: connection established\n",
580 switch (task
->tk_status
) {
583 dprintk("RPC: %5u xprt_connect_status: server %s refused "
584 "connection\n", task
->tk_pid
,
585 task
->tk_client
->cl_server
);
588 dprintk("RPC: %5u xprt_connect_status: connection broken\n",
592 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
593 "out\n", task
->tk_pid
);
596 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
597 "server %s\n", task
->tk_pid
, -task
->tk_status
,
598 task
->tk_client
->cl_server
);
599 xprt_release_write(xprt
, task
);
600 task
->tk_status
= -EIO
;
605 * xprt_lookup_rqst - find an RPC request corresponding to an XID
606 * @xprt: transport on which the original request was transmitted
607 * @xid: RPC XID of incoming reply
610 struct rpc_rqst
*xprt_lookup_rqst(struct rpc_xprt
*xprt
, __be32 xid
)
612 struct list_head
*pos
;
614 list_for_each(pos
, &xprt
->recv
) {
615 struct rpc_rqst
*entry
= list_entry(pos
, struct rpc_rqst
, rq_list
);
616 if (entry
->rq_xid
== xid
)
620 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
622 xprt
->stat
.bad_xids
++;
625 EXPORT_SYMBOL_GPL(xprt_lookup_rqst
);
628 * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
629 * @task: RPC request that recently completed
632 void xprt_update_rtt(struct rpc_task
*task
)
634 struct rpc_rqst
*req
= task
->tk_rqstp
;
635 struct rpc_rtt
*rtt
= task
->tk_client
->cl_rtt
;
636 unsigned timer
= task
->tk_msg
.rpc_proc
->p_timer
;
639 if (req
->rq_ntrans
== 1)
640 rpc_update_rtt(rtt
, timer
,
641 (long)jiffies
- req
->rq_xtime
);
642 rpc_set_timeo(rtt
, timer
, req
->rq_ntrans
- 1);
645 EXPORT_SYMBOL_GPL(xprt_update_rtt
);
648 * xprt_complete_rqst - called when reply processing is complete
649 * @task: RPC request that recently completed
650 * @copied: actual number of bytes received from the transport
652 * Caller holds transport lock.
654 void xprt_complete_rqst(struct rpc_task
*task
, int copied
)
656 struct rpc_rqst
*req
= task
->tk_rqstp
;
658 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
659 task
->tk_pid
, ntohl(req
->rq_xid
), copied
);
661 task
->tk_xprt
->stat
.recvs
++;
662 task
->tk_rtt
= (long)jiffies
- req
->rq_xtime
;
664 list_del_init(&req
->rq_list
);
665 /* Ensure all writes are done before we update req->rq_received */
667 req
->rq_received
= req
->rq_private_buf
.len
= copied
;
668 rpc_wake_up_task(task
);
670 EXPORT_SYMBOL_GPL(xprt_complete_rqst
);
672 static void xprt_timer(struct rpc_task
*task
)
674 struct rpc_rqst
*req
= task
->tk_rqstp
;
675 struct rpc_xprt
*xprt
= req
->rq_xprt
;
677 dprintk("RPC: %5u xprt_timer\n", task
->tk_pid
);
679 spin_lock(&xprt
->transport_lock
);
680 if (!req
->rq_received
) {
681 if (xprt
->ops
->timer
)
682 xprt
->ops
->timer(task
);
683 task
->tk_status
= -ETIMEDOUT
;
685 task
->tk_timeout
= 0;
686 rpc_wake_up_task(task
);
687 spin_unlock(&xprt
->transport_lock
);
691 * xprt_prepare_transmit - reserve the transport before sending a request
692 * @task: RPC task about to send a request
695 int xprt_prepare_transmit(struct rpc_task
*task
)
697 struct rpc_rqst
*req
= task
->tk_rqstp
;
698 struct rpc_xprt
*xprt
= req
->rq_xprt
;
701 dprintk("RPC: %5u xprt_prepare_transmit\n", task
->tk_pid
);
703 spin_lock_bh(&xprt
->transport_lock
);
704 if (req
->rq_received
&& !req
->rq_bytes_sent
) {
705 err
= req
->rq_received
;
708 if (!xprt
->ops
->reserve_xprt(task
)) {
713 if (!xprt_connected(xprt
)) {
718 spin_unlock_bh(&xprt
->transport_lock
);
722 void xprt_end_transmit(struct rpc_task
*task
)
724 xprt_release_write(task
->tk_xprt
, task
);
728 * xprt_transmit - send an RPC request on a transport
729 * @task: controlling RPC task
731 * We have to copy the iovec because sendmsg fiddles with its contents.
733 void xprt_transmit(struct rpc_task
*task
)
735 struct rpc_rqst
*req
= task
->tk_rqstp
;
736 struct rpc_xprt
*xprt
= req
->rq_xprt
;
739 dprintk("RPC: %5u xprt_transmit(%u)\n", task
->tk_pid
, req
->rq_slen
);
741 if (!req
->rq_received
) {
742 if (list_empty(&req
->rq_list
)) {
743 spin_lock_bh(&xprt
->transport_lock
);
744 /* Update the softirq receive buffer */
745 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
,
746 sizeof(req
->rq_private_buf
));
747 /* Add request to the receive list */
748 list_add_tail(&req
->rq_list
, &xprt
->recv
);
749 spin_unlock_bh(&xprt
->transport_lock
);
750 xprt_reset_majortimeo(req
);
751 /* Turn off autodisconnect */
752 del_singleshot_timer_sync(&xprt
->timer
);
754 } else if (!req
->rq_bytes_sent
)
757 status
= xprt
->ops
->send_request(task
);
759 dprintk("RPC: %5u xmit complete\n", task
->tk_pid
);
760 spin_lock_bh(&xprt
->transport_lock
);
762 xprt
->ops
->set_retrans_timeout(task
);
765 xprt
->stat
.req_u
+= xprt
->stat
.sends
- xprt
->stat
.recvs
;
766 xprt
->stat
.bklog_u
+= xprt
->backlog
.qlen
;
768 /* Don't race with disconnect */
769 if (!xprt_connected(xprt
))
770 task
->tk_status
= -ENOTCONN
;
771 else if (!req
->rq_received
)
772 rpc_sleep_on(&xprt
->pending
, task
, NULL
, xprt_timer
);
773 spin_unlock_bh(&xprt
->transport_lock
);
777 /* Note: at this point, task->tk_sleeping has not yet been set,
778 * hence there is no danger of the waking up task being put on
779 * schedq, and being picked up by a parallel run of rpciod().
781 task
->tk_status
= status
;
782 if (status
== -ECONNREFUSED
)
783 rpc_sleep_on(&xprt
->sending
, task
, NULL
, NULL
);
786 static inline void do_xprt_reserve(struct rpc_task
*task
)
788 struct rpc_xprt
*xprt
= task
->tk_xprt
;
793 if (!list_empty(&xprt
->free
)) {
794 struct rpc_rqst
*req
= list_entry(xprt
->free
.next
, struct rpc_rqst
, rq_list
);
795 list_del_init(&req
->rq_list
);
796 task
->tk_rqstp
= req
;
797 xprt_request_init(task
, xprt
);
800 dprintk("RPC: waiting for request slot\n");
801 task
->tk_status
= -EAGAIN
;
802 task
->tk_timeout
= 0;
803 rpc_sleep_on(&xprt
->backlog
, task
, NULL
, NULL
);
807 * xprt_reserve - allocate an RPC request slot
808 * @task: RPC task requesting a slot allocation
810 * If no more slots are available, place the task on the transport's
813 void xprt_reserve(struct rpc_task
*task
)
815 struct rpc_xprt
*xprt
= task
->tk_xprt
;
817 task
->tk_status
= -EIO
;
818 spin_lock(&xprt
->reserve_lock
);
819 do_xprt_reserve(task
);
820 spin_unlock(&xprt
->reserve_lock
);
823 static inline __be32
xprt_alloc_xid(struct rpc_xprt
*xprt
)
828 static inline void xprt_init_xid(struct rpc_xprt
*xprt
)
830 xprt
->xid
= net_random();
833 static void xprt_request_init(struct rpc_task
*task
, struct rpc_xprt
*xprt
)
835 struct rpc_rqst
*req
= task
->tk_rqstp
;
837 req
->rq_timeout
= xprt
->timeout
.to_initval
;
840 req
->rq_buffer
= NULL
;
841 req
->rq_xid
= xprt_alloc_xid(xprt
);
842 req
->rq_release_snd_buf
= NULL
;
843 xprt_reset_majortimeo(req
);
844 dprintk("RPC: %5u reserved req %p xid %08x\n", task
->tk_pid
,
845 req
, ntohl(req
->rq_xid
));
849 * xprt_release - release an RPC request slot
850 * @task: task which is finished with the slot
853 void xprt_release(struct rpc_task
*task
)
855 struct rpc_xprt
*xprt
= task
->tk_xprt
;
856 struct rpc_rqst
*req
;
858 if (!(req
= task
->tk_rqstp
))
860 rpc_count_iostats(task
);
861 spin_lock_bh(&xprt
->transport_lock
);
862 xprt
->ops
->release_xprt(xprt
, task
);
863 if (xprt
->ops
->release_request
)
864 xprt
->ops
->release_request(task
);
865 if (!list_empty(&req
->rq_list
))
866 list_del(&req
->rq_list
);
867 xprt
->last_used
= jiffies
;
868 if (list_empty(&xprt
->recv
))
869 mod_timer(&xprt
->timer
,
870 xprt
->last_used
+ xprt
->idle_timeout
);
871 spin_unlock_bh(&xprt
->transport_lock
);
872 xprt
->ops
->buf_free(req
->rq_buffer
);
873 task
->tk_rqstp
= NULL
;
874 if (req
->rq_release_snd_buf
)
875 req
->rq_release_snd_buf(req
);
876 memset(req
, 0, sizeof(*req
)); /* mark unused */
878 dprintk("RPC: %5u release request %p\n", task
->tk_pid
, req
);
880 spin_lock(&xprt
->reserve_lock
);
881 list_add(&req
->rq_list
, &xprt
->free
);
882 rpc_wake_up_next(&xprt
->backlog
);
883 spin_unlock(&xprt
->reserve_lock
);
887 * xprt_set_timeout - set constant RPC timeout
888 * @to: RPC timeout parameters to set up
889 * @retr: number of retries
890 * @incr: amount of increase after each retry
893 void xprt_set_timeout(struct rpc_timeout
*to
, unsigned int retr
, unsigned long incr
)
896 to
->to_increment
= incr
;
897 to
->to_maxval
= to
->to_initval
+ (incr
* retr
);
898 to
->to_retries
= retr
;
899 to
->to_exponential
= 0;
903 * xprt_create_transport - create an RPC transport
904 * @args: rpc transport creation arguments
907 struct rpc_xprt
*xprt_create_transport(struct rpc_xprtsock_create
*args
)
909 struct rpc_xprt
*xprt
;
910 struct rpc_rqst
*req
;
912 switch (args
->proto
) {
914 xprt
= xs_setup_udp(args
);
917 xprt
= xs_setup_tcp(args
);
920 printk(KERN_ERR
"RPC: unrecognized transport protocol: %d\n",
922 return ERR_PTR(-EIO
);
925 dprintk("RPC: xprt_create_transport: failed, %ld\n",
930 kref_init(&xprt
->kref
);
931 spin_lock_init(&xprt
->transport_lock
);
932 spin_lock_init(&xprt
->reserve_lock
);
934 INIT_LIST_HEAD(&xprt
->free
);
935 INIT_LIST_HEAD(&xprt
->recv
);
936 INIT_WORK(&xprt
->task_cleanup
, xprt_autoclose
);
937 init_timer(&xprt
->timer
);
938 xprt
->timer
.function
= xprt_init_autodisconnect
;
939 xprt
->timer
.data
= (unsigned long) xprt
;
940 xprt
->last_used
= jiffies
;
941 xprt
->cwnd
= RPC_INITCWND
;
942 xprt
->bind_index
= 0;
944 rpc_init_wait_queue(&xprt
->binding
, "xprt_binding");
945 rpc_init_wait_queue(&xprt
->pending
, "xprt_pending");
946 rpc_init_wait_queue(&xprt
->sending
, "xprt_sending");
947 rpc_init_wait_queue(&xprt
->resend
, "xprt_resend");
948 rpc_init_priority_wait_queue(&xprt
->backlog
, "xprt_backlog");
950 /* initialize free list */
951 for (req
= &xprt
->slot
[xprt
->max_reqs
-1]; req
>= &xprt
->slot
[0]; req
--)
952 list_add(&req
->rq_list
, &xprt
->free
);
956 dprintk("RPC: created transport %p with %u slots\n", xprt
,
963 * xprt_destroy - destroy an RPC transport, killing off all requests.
964 * @kref: kref for the transport to destroy
967 static void xprt_destroy(struct kref
*kref
)
969 struct rpc_xprt
*xprt
= container_of(kref
, struct rpc_xprt
, kref
);
971 dprintk("RPC: destroying transport %p\n", xprt
);
973 del_timer_sync(&xprt
->timer
);
976 * Tear down transport state and free the rpc_xprt
978 xprt
->ops
->destroy(xprt
);
982 * xprt_put - release a reference to an RPC transport.
983 * @xprt: pointer to the transport
986 void xprt_put(struct rpc_xprt
*xprt
)
988 kref_put(&xprt
->kref
, xprt_destroy
);
992 * xprt_get - return a reference to an RPC transport.
993 * @xprt: pointer to the transport
996 struct rpc_xprt
*xprt_get(struct rpc_xprt
*xprt
)
998 kref_get(&xprt
->kref
);